This is the first time I've made a GitHub repo public. I'd already hinted in the RSS Club that I've written my own RSS reader, one that uses an LLM to sort feed items by relevance for me. I don't have to read everything that crosses my path. We're pumped full of news like mad these days as it is. An anti-FOMO RSS reader, if you like. The name is still a bit clunky.
The features are modest:
- Feeds are taken from an OPML file and read, and the items are stored in a MySQL table.
- If an
ANTHROPIC_API_KEYis set, newly fetched items are classified using the claude-haiku-4-5-20251001 model. This relies on a Markdown file containing a list of categories, one per line, each with keywords or a definition. The order of the categories reflects their relevance. Anything the LLM can't classify is filed under "Nicht kategorisiert" (uncategorised) and sits at the bottom of the list. Those tend to be posts that fall outside my areas of interest. Classification incurs a small cost. - New items can be marked as read.
- Old items are binned after 5 days.
- Items can be favourited, in which case they're kept beyond the 5 days.
That's all there is so far. I could imagine dropping the index.php for reading altogether and instead serving the whole thing purely as an RSS feed with full posts (where available), which I could then read in NetNewsWire. We'll see …
A DDEV config is included in the repo.
Well, it's more of a proof of concept than a serious application. I suspect you could do something similar in Inoreader with a few clicks. But where's the fun in that?